Classification of muscle activity patterns in healthy children using biclustering algorithm

Abstract

In recent years, there has been major interest in recognising electromyography (EMG) patterns. This work proposes a new method based on a biclustering algorithm which can group strides showing homogeneous EMG activation intervals. The surface EMG signals of biceps femoris, rectus femoris, semitendinosus, lateral gastrocnemius, and medial gastrocnemius muscles of 17 healthy children aged between 4 and 11 years old were obtained using a Trigno EMG wireless system. The data set was tested for different values of parameter α (the threshold describing when the multiple node deletion step is used) and δ (the threshold that limits the value of the mean square residue). The highest number of coincidences of muscle activation was observed in 6 to 7-year-old subjects. This was not affected by their anthropometrics or gender. The obtained biclusters reflect actual differences between the subjects’ gait parameters, namely stride length, stride time, and walking speed. These results can be used to develop strategies for finding homogeneous groups of patients.